Navigation – Plan du site

Agglomeration economies, spatial dependence and local industry growth

Raffaele Paci et Stefano Usai
p. 87-109


L’objectif principal de cet article est d’analyser les déterminants de la croissance de l’emploi industriel au niveau local. Par l’estimation d’un modèle empirique nous évaluons le rôle de différents types d’économies d’agglomération sur la croissance locale en contrôlant la présence de dépendance spatiale. L’étude porte sur 784 « systèmes locaux de travail » en Italie et 34 secteurs (21 industries manufacturières et 13 services) sur la période 1991-2001. Les résultats économétriques montrent que la croissance locale est fortement différenciée selon les secteurs. L’impact positif des externalités de diversité et du capital social et humain ainsi que l’impact négatif des externalités de spécialisation et du niveau de concurrence sont cependant à remarquer. Des phénomènes d’association spatiale sont aussi détectés montrant que la croissance locale au sein d’une aire spécifique bénéficie des performances positives des aires voisines.

Haut de page

Texte intégral

I. — Introduction*

  • * We thank Barbara Dettori and Margherita Meloni for their excellent research assistance. Participant (...)

1Economists and geographers have recently refocused their attention on the contribution of agglomeration economies on the process of local economic growth. This phenomenon entails several issues which have been examined extensively by the economic literature from both a theoretical (Baldwin and Martin, 2004) and an empirical point of view (Rosenthal and Strange, 2004; Combes and Overman, 2004).

2This paper contributes to the empirical stream of literature by assessing the role of the forces which induce firms co-localisation and may generate different growth patterns both for sectors and local areas. It is argued that such patterns have to be studied together with the process of structural change which characterises modern economies. Indeed, these transformations have insightful implications for the analysis of the spatial distribution of economic activities which, in turn, are highly affected by the process of de-verticalisation and de-localisation of mass production industries and the concurrent development and diffusion of service activities. These trends are modifying the economic geography of local production systems as well as the way these are linked to the global economy. Economic landscapes are increasingly being shaped by a complex mixture of forces operating simultaneously worldwide as much as at the regional level and which share a common denominator: the structural shift from manufacturing to services. The main effect of such phenomenon in space is the fact that, on the one hand, urban areas lose manufacturing capacity to become more service oriented and, on the other hand, peripheral areas become more favourable locations both for manufacturing and service.

3The main aim of this paper is to analyse local economic performances, as expressed by employment dynamics, in the service and in the manufacturing sectors. As in most theoretical and empirical literature, we refer not only to the presence of traditional production factors (capital, labour and natural resources) in a given area, but also to several types of « local externalities  ». We focus on the usual distinction among specialisation – Marshall externalities, diversity – Jacobs economies and competition – Porter effects. Moreover, we analyse the potential influence of firm size, population density, human and social capital.

4Another goal of the paper concerns the use of spatial econometric techniques to take into account the possibility of some cross-border externalities, that is, neighbourhood factors affecting local growth. This allows us to avoid considering geographical units as isolated closed economies and therefore placing artificial bounds to agglomeration economies as sometimes occurred in previous literature, especially in the United States (Glaeser et al., 1992; Henderson et al., 1995).

5The empirical analysis is applied to the 784 Local Labour Systems (LLS) in Italy over the period 1991-2001 and 34 manufacturing and services sectors. The present contribution builds upon our previous work on this topic (Usai and Paci, 2003; Paci and Usai, 2005), since it considers the entire market economy over a longer time period while including some important additional factors among the set of explanatory variables.

6The paper is organised as follows. The next section presents the databank along with a descriptive picture of the employment dynamics at the local and sectoral level. In the third section we briefly survey the literature on local agglomeration. In the fourth section the empirical model is presented. In section five the estimation procedures are described while the main econometric results are discussed in section six. In the last section some concluding remarks are proposed.

II. — Data and descriptive analysis

7The empirical analysis is applied to the 784 LLS in Italy identified in the 1991 population census by the national Statistical Office (ISTAT). LLSs are groupings of municipalities showing a high degree of self containment of commuting workers. This fine level of geographical breakdown appears particularly fruitful for the analysis of local growth since it facilitates the identification of the agglomeration forces at the local level and of the spillover effects arising from contiguous areas. As for the sectoral breakdown, we consider 34 sectors at the 2 digit ATECO91 – ISIC3 level: 21 industrial sectors (including building) and 13 service sectors (excluding the public sector and other non-market services).

8The employment dynamics in Italy in the decade from 1991 until 2001 is positive, with a gain of more than one million units of labour and an average annual increase of 0.7% (see Map 1A, next page). In the nineties, employment in Italy was characterised by a clear divide: until 1995 there was a long crisis whilst in the second period Italy experienced a long expansion which allowed the country to move along a convergence path towards European standards, especially in terms of labour participation.

  • 1 Looking at individual LLSs, one can see the presence of idiosyncratic shocks affecting specific sec (...)

9The positive trend in employment for the whole country is confirmed in all geographical areas but for the Islands (Sicily and Sardinia), which report a slightly negative variation of - 0.01% each year. The highest aggregate increase is, on the contrary, achieved in the North East and in the Centre-South, contributing to the reinforcement of the long-lasting gap between the area of North and Centre, which experiences a 0.8% annual average growth rate, and the area of South and Islands which moves at half of that pace with a mere 0.4 % 1.

10Most importantly for the scope of this paper, these aggregate trends conceal opposed patterns for the two macro-sectors (see Maps 1B and 1C, next page): industry shows an average employment fall of 0.19% per year, whilst the service sector increases by 1.5% per year. These divergent performances are obviously related to the process of structural change common to most OECD countries, with manufacturing employment constantly decreasing and service activities becoming the primary source of employment growth. Such a process has many explanations. One is that labour productivity growth may result from an increase in manufacturing output associated to a fall in employment. This, in turn, is the result of many factors: from the process of re-localization of manufacturing where operating costs are lower, to the practice of outsourcing, which implies that industrial firms, in order to improve their core business’ productivity, move auxiliary activities to external service firms (i.e. cleaning, accounting, engineering, marketing, security, etc.). Nonetheless, it should be noticed that relocation of service activity is accelerating too. Internationally tradable sectors, which are ICT intensive and require low levels of face-to-face interaction, are especially likely to move abroad. This affects both low and high skilled jobs, due to the fact that relocation of services occurs mostly between developed countries and is driven not only by cost but also by quality and market access reasons. Italy is not different: the employment growth process is mainly due to the increase of services, whose share of total labour is constantly expanding.

Map 1. Employment Dynamics in LocalLabour Systems in Italy (% annual average growth rate, 1991-2001)

Map 1. Employment Dynamics in LocalLabour Systems in Italy (% annual average growth rate, 1991-2001)

11Looking at the maps we can see a clustering of growing LLSs in the North-East, both in industrial and service sectors. The North-East is a recent story of industrial and service development based on local networks of small and medium firms and plants scattered throughout the area, which follows the model of the « industrial districts ». On the contrary, the North-West, traditionally based on large heavy industries, presents over the nineties the deepest fall in industrial employment compensated by the highest growth in services. The performance of the other macro-regions is always characterised by positive dynamics for the service sector and negative ones for manufacturing. The combination of these two contrasting trends often produces a positive global performance.

12When one looks at individual sectors, as in Table 1, employment dynamics shows as much variation from sector to sector as from one area to another. The best performing sectors within services are, above all, Real estate activities (10% annual average growth rate) and Computer activities (6.7%). Some services have, however, negative dynamics: Retail trade (which has the highest employees’ share ) and Post and telecommunication. The worst performing sectors in manufacturing are primarily Wearing apparel (-3.4%), the upward related sector of Textiles (-2.6%) and also the two transport industries. Few manufacturing sectors show a positive performance: Rubber and plastic (+1.9%), and Fabricated Metal products (+1,3%). The employment dynamics in the Building sector is also positive (1.4%).

  • 2 The Moran index is based on the notion of binary contiguity between spatial units, where the generi (...)

13Table 1, next page reports also the Moran test for spatial dependence in the employment growth rates among the 784 LLSs2. At the global and macro-sector level we find evidence of spatial autocorrelation, already perceivable from the visual inspection of the previous maps. The Moran index for the whole country and for the industrial and service sectors indicates that employment dynamics in an LLS is influenced by the performance of nearby areas. At the same time, when employment growth is disaggregated by sector, the occurrence of spatial dependence also shows sectoral differences. As an example, one can see in Map 2, next page the contrasting patterns of Wearing apparel where employment dynamics clearly show a tendency to cluster in contiguous areas; and those of Chemicals which are more scattered over space. Spatial association is, therefore, found only in the former case, while in the latter the index is not significant. More generally, a positive and significant spatial dependence is found in 21 sectors out of 34 (12/21 in industry and 9/13 in services). In one sector, Coke and petroleum products, a negative significant spatial dependence is detected, due to the extremely dispersed and polarised structure of this industry.

Table 1. Employment growth and spatial autocorrelation at the sectoral level, 1991-2001

Table 1. Employment growth and spatial autocorrelation at the sectoral level, 1991-2001

14In conclusion, two main findings emerge from the descriptive analysis of employment dynamics. Firstly, a highly differentiated pattern among sectors; secondly, the presence of spatial association among contiguous areas in the employment growth rates. In the econometric estimation these stylised facts are directly taken into account.

III. — Literature background

15In the last decade, the influence of agglomeration and other forms of local externalities on economic growth has been under recurrent investigation. Glaeser et al. (1992) wrote the first contribution to focus on employment growth as a proxy for local economic performance and to study its dynamics at both the city and the sectoral level. Their empirical analysis is based on the distinction between static externalities, associated with cost efficiencies, and dynamic externalities, related to knowledge spillovers. Both types of externalities are potentially related to Marshall-Arrow-Romer (MAR) localisation economies which encourage growth via industrial specialisation (Marshall, 1890; Arrow, 1962; Romer, 1986). In fact, Glaeser et al. (1992) included also Porter’s (1990) and Jacobs’s (1969) hypotheses that competition and industrial diversity, respectively, enhance local growth. Since then, the debate about dynamic externalities has mainly focused on three competing theories which have different predictions on the relationship between agglomeration phenomena and economic growth.

16The main difference between these theories concerns the effects of industrial specialisation (the degree to which a location specializes in one industry), diversity (the variety of sectors in a location) and competition (the local market structure) (see Van Oort, 2004). The MAR framework maintains that most spillovers occur among firms within the same industry; specialized locations with high levels of industry concentration should, therefore, experience more innovation and faster growth. In contrast, Jacobs suggests that positive externalities arise across different industries and predicts that industries innovate more and grow faster in locations with greater diversity and competition. Porter accepts the idea of localisation economies but, in accordance with Jacobs, asserts that local competition induces more innovation and hence local economic growth.

Map 2. Employment growth rate and spatial association in selected sectors (% annual average, 1991-2001)

Map 2. Employment growth rate and spatial association in selected sectors (% annual average, 1991-2001)

17Empirical tests addressing this debate have produced conflicting results depending on the proxy of economic growth used as the dependent variable (employment, labour productivity, TFP, wages), on the level of geographical aggregation (firms data, local systems, metropolitan areas, regions) and on the sectoral level (2- or 3- digit, employment, services). Glaeser et al. (1992) find that both competition and diversity foster industry growth measured by employment dynamics, while specialization discourages it. As for Porter effects, the indicator of local market structure (the firm dimension) is unsatisfactory. The evidence collected for other countries, mainly in the European Union, seems to support these findings. For example in the case of Italy, Usai and Paci (2003) find a positive effect on growth played by diversity and a negative one by specialization at the local labour system level. In the Netherlands, Van Soest et al. (2006) find similar results at the city level.

  • 3 These outcomes (as the one in Glaeser et al., 1992) is subject to Combes’ critique (2000a), accordi (...)

18These results conflict with those of Henderson et al. (1995) who reported positive effects for both diversity and specialization externalities for high tech industries while finding only MAR spillovers for mature industries. The importance of taking into account all relevant differences across sectors is raised also by Forni and Paba (2002), who find that in Italy specialization and variety matter for growth in most manufacturing sectors, even though they show that each industry needs its own variety in terms of input-output relations 3.

19Along this research path, Maggioni (2002), concentrates on high tech sectors in four countries (US, UK, France and Italy), and finds that geographical specialisation may prove crucial together with a competitive environment. In this vein Combes (2000b) for France, Paci and Usai (2005) for Italy and Blien and Suedekum (2005) for Germany have tried to improve the analysis by examining both the manufacturing and the service sector. Although a positive effect is generally associated with diversity and a negative one with specialization, such a choice is insightful given that these externalities prove to be rather different across sectors. As far as the Porter effect is concerned, this is correctly measured by Combes (2000b) who finds a negative effect on growth in industry and no significant effect in services.

  • 4 He also studies the spatial range of externalities and finds that they are quite localized within t (...)

20The contributions of Dekle (2002) and Cingano and Schivardi (2004) are particularly important because they remark that employment growth cannot be used as a valid proxy of productivity growth unless four conditions hold. These conditions concern the constancy of local capital stock, the demand elasticity, the effects of agglomeration on labour supply and the degree of substitutability among factors. In the light of such considerations, Cingano and Schivardi (2004) for Italy, and more recently Almeida (2005) for Portugal, show that within the same sample, results are different when employment growth is substituted as dependent variable by other more correct proxies for economic growth (TFP and wages). In fact, specialization effects became positive. Similar results are found by De Lucio et al. (2002), who report no effect of diversity on labour productivity growth and an interesting U-shaped curve for specialisation effects. Finally, Henderson (2003), through the estimation of plant level production functions in a panel context, finds that MAR externalities have strong productivity effects in high-tech but not in machinery industries. He also finds no evidence of urbanization economies from the diversity of local economic activity outside the industry under analysis and limited evidence of urbanization economies4 altogether.

21The use, as the dependent variable, of either TFP or labour productivity is an obvious notable improvement by these studies because this allows to measure in a more accurate way the economic performance of firms and areas. For instance, a highly specialised industrial district may follow a reorganization process based on the reduction of employment which may end up with an increase of the economic performances. However, the choice of productivity measures often implies some inconvenience in terms of data availability. In particular, Dekle (2002) and De Lucio et al. (2002) have to move to a more aggregated geographical level (i.e., administrative regions) where the effects of local externalities are difficult to assess. On the contrary, Henderson (2003) and Cingano and Schivardi (2004) are able to keep a very disaggregated spatial level of analysis at the cost of relying on samples of plant data for the calculation of the productivity measures which bring about serious problems of selection bias.

22Another interesting issue raised in the literature is whether the role of externalities varies with some concurrent economic phenomena. Glaeser et al. (1992), for instance, suggested that there might be an industry life cycle in which externalities are only important in the early development stages. Similarly, Krugman (1991) indicates that as an industry develops, it might become less dependent on pooled labor, specialized inputs and knowledge spillovers. Moreover, externalities that foster the initial development of a location might not be the same that affect its subsequent growth (Duranton and Puga, 2001). In other words, the nature of externalities is not independent from product cycle: experimental activities are initially found in large diverse urban areas (Jacobs externalities) with a large number of small economic enterprises (Porter externalities). Conversely, traditional productions, which are more standardized, can be more decentralized in smaller specialized urban areas with lower costs (Marshall externalities) where large plants can operate more efficiently.

23Most importantly for the purposes of this paper, the role of externalities may vary across industries and, most of all, between the two macro-sectors: manufacturing and service. The reason is, as argued by Krugman and Venables (1995), that goods which are essentially non-tradable (such as services) have to be produced close to customers, leading activities to remain spread out. On the contrary, tradable goods, such as manufacturing, can enjoy agglomeration economies by locating where it is more convenient and therefore more concentrated in space. This view, according to Desmet and Fafchamps (2003), may have interesting dynamic implications. As transport costs fall, goods became tradable, allowing production to take advantage of agglomeration economies in the first place. Later, however, if transport costs continue to drop, those agglomeration economies may go beyond a threshold where activities start spreading back out to less congested areas.

24Finally, the dynamics of the service sectors is linked to the evolution of the manufacturing sector through two connections. First, an inverse relationship between the two macro-sectors, resulting from the fact that manufacturing firms increasingly outsource part of their activities to service firms. Second, a positive relationship because the manufacturing sector is a buyer of services and the two macro-sectors are thus complementary. However, one should bear in mind that service sectors are extremely heterogeneous: for example business services may follow an altogether different localisation process with respect to personal services. Business services are, on average, agglomerated close to the firms to which they sell their products. This is usually explained by referring to intangible aspects of localised knowledge which needs daily and face to face contacts to facilitate exchanges of essential information. Conversely, personal services are usually more spread out in the territory. The complexity of the nature of the two macro-sectors and of their relationship is bound to be reflected in our results, especially those devoted to single sector analysis.

IV. — The empirical model

25Factors affecting employment dynamics of a certain industry in a given area can be decomposed into three major groups: 1.) local and industry specific level, 2.) local specific level, 3.) industry specific level. We will now discuss in some details the various phenomena which have a potential effect on the performance of local sectoral employment (see the Appendix for the variables’ description and sources).

Local and industry specific level

26The first group of determinants are those which are considered idiosyncratic for both area and industry. They include the three types of externalities (specialisation, diversity, market power) previously discussed in the survey of the literature.

27Specialisation externalities (SE), measured by an index of relative production specialisation (the widely used comparative advantage index), should capture the consequences of producing in an area crowded with firms which belong to the same sector. This variable covers both static and dynamic externalities. The former include pecuniary and localisation externalities such as the availability of suitable supplies of labour force, primary and intermediate goods (Ellison and Glaeser, 1999), the provision of specific goods and services (Bartelsman et al., 1994) and the availability of specific infrastructures and networks. The latter take into account dynamic spillovers coming from the intra-industry flows of localised knowledge which occur among similar firms located in the same area (Henderson et al., 1995; see Audretsch and Feldman, 2004 for an up to date survey).

28Diversity externalities (DE) are measured by the inverse of the Herfindhal index applied to employment in all sectors except the one considered. Such externalities are expected to positively influence local growth under the hypothesis that a firm located in a certain area can benefit from the presence in the same area of a wide range of other firms operating in different sectors since it can enjoy profitable inter-industries interactions and cross fertilisation. Furthermore, diversity can be seen as a portfolio strategy which protects the regional economy, and especially local employment, from sector specific shocks (Frenken et al., 2007).

29The degree of market power (MP) is assessed by means of an Herfindhal index based on employees distribution over plants following Lafourcade and Mion’s (2007) methodology. The predicted effect is uncertain, since we may have a positive effect on employment dynamics coming from a higher incentive to innovate for firms with some market power but also a negative one coming from demand pressures on inputs and intermediate goods.

30Finally, we consider firm size (FS) measured in number of employees, which may affect labour dynamics either because of internal economies of scale or because of different dynamics towards optimal size by firms of different dimensions (O’hUallachàin and Satterthwaite, 1992). This way we are able to distinguish between the two effects – competition and scale economies – defining two different indicators and including both of them in the estimated equation.

Local specific level

31Employment changes at the local industry level may be due to some features which characterise the whole local labour system. Local factors may refer to a large set of socio-economic phenomena which influence firms’ performance in the area.

32The first variable focuses on the supply side by taking into account the presence of small firms (SF) within the local economy. The idea is that a larger share of small plants may induce firms to find their optimal production scale through cooperation and integration with other firms at the local level while stimulating the creation of local network externalities (Brusco, 1982). The opposite happens with large firms which are more vertically integrated and therefore are less involved in local networks.

33The second variable takes into account the influence of the size of the local economic system, measured by the population density (PD) in each LLS where a firm is located. A positive effect on local growth is expected when a larger population density implies, among others, a higher and differentiated local demand as well as the availability of a wider supply of local public services (Ciccone and Hall, 1996). Proximity to buyers may have both a static and a dynamic effect, the latter being related to the fact that this may facilitate early perception of market needs. At the same time, increases in the size of the local economy may imply congestion effects giving rise to pollution and other local diseconomies.

34The role of human capital (HK) in facilitating innovation activities, information spillovers and therefore growth is considered by looking at the presence of labour forces with high levels of education in the local area, measured by the share of population with a university education. The hypothesis is that a higher availability of well educated labour forces represents an advantage for the localization of firms, thus fostering local growth (Rauch, 1993; Moretti, 2004).

35Another important local element which may encourage innovation activities and smooth the process of knowledge diffusion is social capital (SK). It is not an easy task to find the proper indicators for such a complex and intangible phenomenon (Helliwell and Putnam, 1995). In this paper we are interested to measure the degree of trust in the local society which may favour firms productivity. Thus we include an index of the propensity to cooperate among firms, based on the number of inter-firms agreement and participations in consortia surveyed by the industrial census at the provincial level. The assumption is that a higher degree of propensity to cooperate among firms in a certain area helps local growth since it facilitates knowledge diffusion, decreases transaction costs and enables firms to take advantage of local externalities.

36Finally, we follow the idea that externalities may affect labour supply (Cingano and Schivardi, 2004) and therefore we try to include this potential effect by inserting an indicator of the labour force size (LS) computed as the share of labour forces over population age 15-65.

Industry specific level

37The growth rate of employment in a local industry may also be affected by factors which are peculiar to each production sector but common to all geographical areas. These factors can capture, for instance, different levels of technological opportunities within each industry at the national level or sectoral demand shocks. They are proxied by the 34 sectoral fixed effects (FE) in the regressions, where the two dimensions (geographical and sectoral) are considered together in a panel framework, while they are, obviously, redundant in the sectoral estimates.

V. — The estimation procedure

38We attempt to consider simultaneously different factors which potentially affect local employment dynamics. The same general specification is applied to sub samples to establish if there is any difference in the value, sign and significance of the estimated coefficients and thus in the role of the explanatory factors. The econometric analysis is based on a simple reduced form where the employment growth rate (gLij), which refers to the local labour system i and the industry j, is affected by the three sets of phenomena described in the previous section:


gLijt+1= β1SEijt+ β 2DEijt+ β 3MPijt+ β 4FSijt+δ1 SFit +δ 2 PDit +δ 3 HKit +δ 4 SKit+δ 5 LSit+ ΣγjFEj+ εij

  • 5 Consequently, from the total number of 26,656 potential observations (given by 784 LLSs x 34 sector (...)

39It is important to remark that all our regressors are exogenous to the local industry employment growth rate (1991-2001) since they refer to the beginning of the period considered (1991). We have excluded local industry observations with zero employee in the initial year (4,373 cases) due to the impossibility of calculating the growth rates and also the observations with zero employee in the final year (939 cases) because this gives rise to extreme growth rate values with typical outlier characteristics5.

Spatial analysis

  • 6 In the estimations for the whole economy with both the geographical and sectoral dimensions it is n (...)

40As we have remarked before, one interesting feature in the analysis at the local level is that employment growth in a region may be influenced by factors which are outside the region. In other words, the economic environment in the nearby areas may influence employment dynamics in a given region and this introduces a possible bias in the regression analysis if we do not control for this element. In order to deal with the problem of spatial association in the sectoral regressions6 we apply the following estimation procedures.

41i. OLS estimation to assess the presence of first order spatial autocorrelation based on the Kelejan and Robinson test which, in contrast to the Moran’s I and Lagrange Multiplier tests, does not require normality of the error term (a problem which is pervasive in our regressions).

42ii. If there is no autocorrelation, then the least squares estimates are efficient and consistent. In such a case, equation 1 is estimated by the OLS White-robust estimation which allows to take into account potential heteroskedasticity, which is found in several cases.

  • 7 In the former case, which is the most relevant, spatial dependence implies the presence of an inter (...)

43iii. If spatial dependence is detected, we draw on LM tests, despite they can be problematic with a-normal errors, in order to assess its possible form: substantive or nuisance7. In both cases the Maximum Likelihood (ML) estimation is usually suggested instead of OLS. However, this is not appropriate in this case, given that this method is sensitive to deviations from normality and homoscedasticity. Consequently, we apply General Method of Moments (GMM) in the nuisance case with spatial errors and Instrumental Variables (IV) in the substantive case with spatial lags.

44iv. In the substantive case, a spatial lag of the dependent variable is included up the contiguity level necessary to correct for the presence of spatial autocorrelation.

45v. In the nuisance model, spatial association is accounted for by estimating the spatial autoregressive coefficient λ for the error lag.

VI. — Econometric results

46Table 2 (bellow) reports the aggregate estimations based on a dataset with two-dimensions (geographical and sectoral) and with cross section weights and fixed effects to control for sectoral differences, while the results of sectoral regressions are reported in Table 3 (bellow). Let us now discuss the main results for each explanatory variable.

Table 2. Econometric results for whole economy and macro-sectors

Table 2. Econometric results for whole economy and macro-sectors

Dependent variable: employment growth in the local industry, annual average 1991-2001.
Estimation method: GLS with cross section weights and White robust standard errors.
Panel estimation by LLS and sectors and with 34 industry Fixed Effects.
Level of significance: ***= 1%; **= 5%; *= 10%.
Note: we have excluded local industry with zero employee in 1991 or 2001

Table 3. Econometric results for sectors

Table 3. Econometric results for sectors

Dependent variable: employment growth in the local industy. Annual average 1991-2001 - Cross-section estimation by LLS - Level of significance: ***= 1%; **= 5%; *= 10%
Estimation Method: OLS-W: Ordinary Least Squares Estimation-White robust Standard Error, GM-error: General method of moments, IV-lag: Instrumental Variables; constant is included
Note: we have excluded local industry with zero employee in 1991 or 2001

47Specialisation externalities. The first important result is the absence of positive specialisation externalities: the coefficient of SE is negative and significant in all panel estimations and in all service sectors. This outcome confirms previous studies for the US (Glaeser et al., 1992), France (Combes, 2000b) and Italy (Cunat and Peri, 2001, Forni and Paba, 2000; Usai and Paci, 2003). One possible explanation for the negative effect of specialisation in services is the fact that probably these sectors are becoming more and more diffused across space closer to consumers. At the same time it is possible that services, thanks to technological change which is labour saving, is substituting capital for labour implying a reduction of employment and a potential increase in productivity. Specialisation has no effect in most manufacturing sectors, since coefficients are often not significantly different from zero, whilst it has a positive effect in two of the most traditional sectors, Wearing apparel and Leather and footwear. These two sectors are among those which are losing more employees and it seems that this global dynamics is stronger in those areas where specialisation was relatively weak, whereas local specialised districts manage to preserve some of the past strength in terms of work force.

48Diversity externalities. We find robust evidence for the positive and significant role played by diversity externalities (DE) both for the panel estimates and in most sectors especially in the services activities (9 out 13). This may be due to several externalities at work. On the one hand, pecuniary externalities due to the fact that firms benefit from the presence of a wide ranging local availability of supply and demand linkages. On the other hand, externalities may be due to knowledge spillovers which move across sectors. Moreover, additional indicators should be computed for this variable in order to disentangle those effects which are truly cross-fertilisation spillovers (and therefore more dynamic in nature) from those which are due to input-output relationships (and therefore with more static consequences). Finally, it would also be interesting to investigate further the role of diversity as a portfolio strategy which defends local areas from sudden demand and supply shocks which hit only some sectors.

49Market power. The market power index (MP) is mostly positive and significant (except for two service sectors) signalling that a competitive environment discourages growth of the local industry and that Porter’s idea does not apply to employment dynamics in Italy. This effect is present both in manufacturing sectors (16 out of 20) and in the service sectors (7 out of 13). Market power, that is low competitive environment, may have positive effects on firms’ ability to grow, thanks to extra-profits which can be invested in innovation (as in the MAR view). Moreover, under the hypothesis of local and segmented markets for immobile inputs (such as land but also skilled labour in Italy) market power can decrease local price competition on inputs favouring employment growth.

50Firms size. As far as the average firm size (FS) is concerned, results show that there is a negative influence of economies of scale both in the global regression and in most sectors. We should remember that this negative influence is not to be interpreted necessarily as the absence of internal economies of scale since we are not measuring effects on productivity but on employment. One way to interpret this result is suggested by Combes (2000b) who finds the same result for France. Since average size is referred to the initial period one may think that the presence of small firms is bound to increase employment growth given that these firms grow faster than big firms which are closer to their optimal dimension. Moreover, knowledge spillovers and flexibility may be higher in small firms, which are therefore better able to adapt to difficult periods, such as the nineties. Interestingly, a positive and statistically significant role is found only inRetail trade, where the last decade has been characterised by a large increase of the plant size.

51Local specific variables. Turning to local specific determinants of employment growth we find a positive influence of the presence of small firms (SF) in the global regressions and in some sectors. These results are in accordance with the Italian production structure characterised by systems of small and medium sized firms with a higher degree of flexibility and ability to respond to the economic cycle.

52The size of the local system, measured by population density (PD), is statistically significant in the aggregate sample and in the service macro-sector, where local demand and urbanisation economies are more important. On the contrary, the estimated coefficients are not significant in the manufacturing macro-sector and in almost all sectoral estimates.

53The indicators referring to the different qualities of capital (human, social) show interesting and composite results. First, university education (HK) emerges as a relevant and positive determinant of local growth (Lodde, 2000) for the whole country, for the service macro-sector and for most service sectors (11 out of 13), but not for manufacturing. On the contrary, several coefficients are not statistically significant or show negative signs in the manufacturing sectors, indicating a less strategic role played by well educated labour forces in the Italian industry, mainly specialised in traditional productions.

54Similarly, social capital, represented by the variable which measures cooperation among firms (SK), is positive and statistically significant, as expected, in the whole country and in the service sectors on the whole but it is not statistically significant in manufacturing. At the sectoral level the effect of social capital is in general not statistically significant in manufacturing while positive and significant values are detected in six service sectors.

55Finally, the presence of a large labour supply (LS) exerts a positive influence on employment dynamics in all aggregate regressions and also in some sectoral estimates.

  • 8 We have also used the distance matrix and main results are maintained.
  • 9 Actually, since the indication of the LM test may be contradictory with anormal errors every model (...)

56Spatial association. The last but not less important result to remark is that, on the basis of the Kelejan and Robinson test, spatial autocorrelation is detected in 20 out of 34 sectors using a first order spatial contiguity matrix8. Therefore either a IV or a GMM estimation is performed according to the indication of the LM tests. More precisely, we implement a lag dependence model (estimated with IV) in six sectors and an error model (estimated with GMM) in 14 sectors9. Thanks to this procedure, spatial autocorrelation has been controlled in all sectors at the first order contiguity. The spatial lagged variables and the spatial autoregressive coefficients have proved positive and significant in 14 sectors, which implies that some form of positive spatial dependence among contiguous areas in employment dynamics is present. In three further sectors (Chemicals, Coke and Renting of machinery), as expected from the Moran analysis, the spatial coefficients are negative and significant, implying that production districts of this type are very polarised. As for the comparison of manufacturing and service sector, we find that spatial association is equally distributed across sectors.

VII. — Concluding remarks

57This paper tries to put the issue of local economic performance in the scenario of ongoing structural change which is transforming modern economies from manufacturing to service ones. It is argued that such a process has insightful implications for the analysis of the geography of economic activities. The main contribution of this paper is, therefore, the analysis of local economic performance, as expressed by employment dynamics, both in the service and in the manufacturing sectors. Thanks to a large set of variables and data we attempt to explain some of the differences in the economic performance of industries by assessing the role of several potential determinants of local economic dynamics.

58Results confirm the existence of a multifaceted picture when it comes to agglomeration forces operating at local geographical level. Overall, we find that specialisation has negative effects on employment growth, possibly due to a process of reorganization which substitute labour with capital and other factors but also due to the diffusion process of services across space. This suggests that local employment, especially in industrialised countries, can benefit from a local production system characterised by a diversified network of small flexible firms willing to cooperate and employing well educated labour forces. Moreover, the presence of some market power seems to imply either less competition for inputs or more resources (due to extra-profits) for investments and therefore positive effects on firm growth. Finally, the presence of spatial association indicates that the growth process in a specific area benefits from the positive performance of the surrounding regions.

59Further, we find evidence of remarkable differences in the influence of local specific externalities between manufacturing and services. As a result of the structural changes, we observe a decline in manufacturing and an increase in services and these opposed dynamics appear to be led by different factors. In the former case, a process of delocalisation is undergoing and this implies the transfer of manufacturing activities out of industrial districts with a reduction of employment. The analysis shows that such industrial districts do not seem to be able any more to exploit the standard set of externalities typical of their success: a collection of specialized small firms based on social networks and human capital. In the latter case of service activities, specialisation externalities have a negative impact since these sectors are becoming more pervasive across space. In spite of that, they remain mainly located in urban areas, where intangible factors (human and social capital) are more relevant to enhance growth within a production environment characterised by a diversified mixture of activities. All these results can have interesting policy implications.

60Finally, it is worth noting that these results somewhat contrast with those obtained by Paci and Usai (2000) for Italy and Moreno et al. (2005) for Europe, whose work investigates on the impact of agglomeration economies on local innovative performances. They find that specialisation rather than diversification externalities are important in the localisation of innovative activity. Our interpretation is that in the developed world specialisation economies, providing they are related to pecuniary externalities, no longer reinforce local industrial districts but only local technological enclaves, as long as such enclaves are mainly due to pure technological spillovers. In other words, factor costs appear to be the main determinant of localisation strategies by firms, as they affect productive structures, while knowledge spillovers are pivotal in innovative performances.

61Differentiated local production systems, together with specialised scientific and technological areas, appear to be the best way to support both employment dynamics and innovative performance in the new century.

Haut de page


Almeida R. (2005), « Local economic structure and growth », Policy Research Working Paper, Series 3728, The World Bank.

Anselin L. and R. Florax (1995), « Advances in Spatial Econometrics ». Berlin: Springer.

Arrow K.-J. (1962), « The Economic Implications of Learning by Doing », Review of Economic Studies, 29, 155-173.

Audretsch D. and M. Feldman (2004), « Knowledge spillovers and the geography of innovation », in Henderson V. and J.-F. Thisse (eds.), « Handbook of Regional and Urban Economics: Cities and Geography ». Amsterdam: Elsevier.

Baldwin R.-E. and P. Martin (2004), « Agglomeration and regional growth », in Henderson V. and J.F. Thisse (eds.), « Handbook of Regional and Urban Economics: Cities and Geography ». Amsterdam: Elsevier.

Bartelsman E., R. Caballero and R. Lyons (1994), « Customer- and Supplier- Driven Externalities », American Economic Review, 84, 1075-1084.

Blien U. and J. Suedekum (2005), « Local Economic Structure and Industry Development in Germany », 1993-2001, Economics Bulletin, 15, 1-8.

Brusco S. (1982), « The Emilian Model: Productive Decentralisation and Social Integration  », Cambridge Journal of Economics, 6, 167- 184.

Ciccone A. and R. Hall (1996), « Productivity and the density of economic activity », American Economic Review, 86, 54-70.

Cingano F. and F. Schivardi (2004), « Identifying the Sources of Local Productivity Growth », Journal of the European Economic Association, 2, 720-742.

Combes P. (2000a), « Marshall-Arrow-Romer Externalities and City Growth », CERAS wp 99/06.

Combes P. (2000b), « Economic Structure and Local Growth: France, 1984-1993 », Journal of Urban Economics, 47, 329-355.

Combes P. and H. Overman (2004), « The Spatial Distribution of Economic Activities in the European Union », in Henderson V. and J.-F. Thisse (eds.), « Handbook of Regional and Urban Economics: Cities and Geography ». Amsterdam: Elsevier.

Cunat A. and G. Peri (2001), « Job creation: Geography, Technology and Infrastructures  », Giornale degli Economisti, 61, 23-54.

De Lucio J.-J, J.-A. Herce and A. Goicolea (2002), « The Effects of Externalities on Productivity Growth in Spanish Industry », Regional Science and Urban Economics, 32, 241-258.

Dekle R. (2002), « Industrial Concentration and Regional Growth: Evidence from the Prefectures », Review of Economics and Statistics, 84, 310-315.

Desmet K. and M. Fafchamps (2003), « What are falling Transport Costs doing to Spatial Concentration Across US Counties? », CEPR Discussion Paper 3853.

Duranton G. and D. Puga (2001), « Nursery Cities. Urban Diversity, Process Innovation, and the Life Cycle of Products », American Economic Review, 91, 1454-1477.

Ellison G. and E. Glaeser (1999), «  The Geographic Concentration of Industry: Does Natural Advantage Explain Agglomeration? », American Economic Review Papers and Proceedings, 89, 301-316.

Forni M. and S. Paba (2000), « The sources of local growth: evidence from Italy », Giornale degli Economisti e Annali di Economia, 59, 1-49.

Forni M. and S. Paba (2002), « Spillovers and the Growth of Local Industries », Journal of Industrial Economics, 50, 151-171.

Frenken K., F.-G. van Oort and T. Verburg (2007) ,« Related variety, unrelated variety and regional economic growth », Regional Studies, 41, 685-697.

Glaeser E.-L., H.-D. Kallal, J.-A. Sheinkman and A. Sheifler (1992), « Growth in Cities », Journal of Political Economics, 100, 1126-1152.

Helliwell J. and R. Putnam (1995), « Economic growth and social capital in Italy  », Eastern Economic Journal, 21, 295-307.

Henderson V. (2003), « Marshall’s scale economies », Journal of Urban Economics, 53, 1-28.

Henderson V., A. Kunkoro and M. Turner (1995), « Industrial Development in Cities  », Journal of Political Economics, 103, 1067-1090.

Jacobs J. (1969), « The Economy of Cities ». London: Jonathan Cape.

Krugman P. (1991), « Economic Geography and Trade ». Cambridge, MA: MIT Press.

Krugman P. and T. Venables (1995), « Globalisation and the inequality of nations  », Quarterly Journal of Economics, 110, 857-880.

Lafourcade M. and G. Mion (2007), « Concentration, agglomeration and the size of plants  », Regional Science and Urban Economics, 29, 575-604.

Lodde S. (2000), « Education and Growth: Some Disaggregate Evidence from the Italian Regions », Rivista Internazionale di Scienze Economiche e Commerciali, 47, 267-293.

Maggioni M. (2002), « Clustering Dynamics and the Location of High-Tech Firms », Heidelberg: Physica-Verlag.

Marshall A. (1890), « Principles of Economics ». London: Macmillan.

Moreno R., R. Paci, S. Usai (2005), « Spatial spillovers and innovation activity in European Regions », Environment and Planning A, 37, 1793-1812.

Moretti E. (2004), « Workers’ Education, Spillovers, and Productivity: Evidence from Plant-Level Production Functions », American Economic Review, 94, 656-690.

O’ hUallachàin B. and M. Satterthwaite (1992), « Sectoral growth patterns at the metropolitan level: an evaluation of economic development incentives », Journal of Urban Economics, 31, 25-58.

Paci R. and S. Usai (2000), « Technological enclaves and industrial districts. An analysis of the regional distribution of innovative activity in Europe », Regional Studies, 34, 97-114.

Paci R. and S. Usai (2005), « Spatial externalities and local employment dynamics », in Vertova G. (ed.), « The Changing Economic Geography of Globalization ». London: Routledge.

Porter M. (1990), « The Competitive Advantage of Nations ». New York: Free Press.

Rauch J. (1993), «  Productivity Gains from Geographic Concentration of Human Capital: Evidence from the Cities », Journal of Urban Economics. 34, 380-400.

Romer P. (1986), « Increasing Returns and Long-run Growth », Journal of Political Economy, 94, 1002-1037.

Rosenthal S. and W. Strange (2004), « Evidence on the nature and sources of agglomeration economies », in Henderson V. and J.-F. Thisse (eds.), « Handbook of Regional and Urban Economics: Cities and Geography ». Amsterdam: Elsevier.

Usai S. and R. Paci (2003), « Externalities and local economic growth in manufacturing industries », in Fingleton B. (ed.), « European Regional Growth ». Berlin: Springer-Verlag.

Van Oort F.-G. (2004), « Urban growth and innovation. Spatially bounded externalities in the Netherlands ». Aldershot: Ashgate.

Van Soest D.-P., S.-D. Gerking and F.-G. van Oort (2006), « Spatial Impacts of Agglomeration Externalities », Journal of Regional Science, 46, 881-899.

Haut de page


Variables description and sources

Variables description and sources
Haut de page


* We thank Barbara Dettori and Margherita Meloni for their excellent research assistance. Participants at ERSA conference in Amsterdam, at the CEPR workshop on Agglomeration Economies and Economic Growth in Cagliari and at the International workshop on Spatial Econometrics and Statistics in Rome have given useful suggestions. We also thank Frank van Oort and an anonymous referee for helpful comments on a previous version of the paper. This research has been performed within the project IAREG, Intangible Assets and Regional Economic Growth - EU 7th Framework Programme Theme 8, SSH-2007-1.1.2

1 Looking at individual LLSs, one can see the presence of idiosyncratic shocks affecting specific sectors prevalent in certain areas. Therefore among the top ten best performing LLSs we find areas both in the South (mainly in Puglia and Basilicata) and in the North (especially in Veneto). For instance, the highest employment growth is recorded at Melfi, in the southern region of Basilicata, where the multinational car maker Fiat built a plant to exploit the financial and fiscal incentives available to the Objective 1 regions of the EU.

2 The Moran index is based on the notion of binary contiguity between spatial units, where the generic element of the first order spatial matrix takes value 1 if the two regions share a common border or 0 otherwise. This contiguity matrix is also used in the econometric estimation. The Moran test can be computed also for higher order contiguity matrices and for the distance matrix. In the latter case the generic element is the inverse of the geographical distance in kilometers between the centroids of regions i and j.

3 These outcomes (as the one in Glaeser et al., 1992) is subject to Combes’ critique (2000a), according to which the simultaneous inclusion of a specialisation index and total employment among the regressors introduces a positive bias on the specialisation coefficient. The positive effect of specialisation is therefore questioned.

4 He also studies the spatial range of externalities and finds that they are quite localized within the same county, so that there are no external benefits from plants in other counties in the MSA.

5 Consequently, from the total number of 26,656 potential observations (given by 784 LLSs x 34 sectors) we have excluded 5,312 observations in various sectors ending up with a total number of 21,344 cases used in the econometric analysis.

6 In the estimations for the whole economy with both the geographical and sectoral dimensions it is not feasible to deal with the problem of spatial association due to technical limits imposed by both Spacestat and Matlab for such large datasets with an unbalanced panel due to the presence of several missing values.

7 In the former case, which is the most relevant, spatial dependence implies the presence of an interactive process among spatial units, whereas in the latter case, dependence is probably a by-product of measurement errors for observations in contiguous spatial units. In the case of spatial error autocorrelation, OLS coefficients estimates are unbiased but inefficient; while in the presence of spatial lag dependence, OLS parameters become biased and inconsistent (Anselin and Florax, 1995).

8 We have also used the distance matrix and main results are maintained.

9 Actually, since the indication of the LM test may be contradictory with anormal errors every model is estimated with both methods (IV and GMM). Most results are maintained.

Haut de page

Table des illustrations

Titre Map 1. Employment Dynamics in LocalLabour Systems in Italy (% annual average growth rate, 1991-2001)
Fichier image/jpeg, 132k
Titre Table 1. Employment growth and spatial autocorrelation at the sectoral level, 1991-2001
Fichier image/jpeg, 132k
Titre Map 2. Employment growth rate and spatial association in selected sectors (% annual average, 1991-2001)
Fichier image/jpeg, 104k
Titre Table 2. Econometric results for whole economy and macro-sectors
Légende Dependent variable: employment growth in the local industry, annual average 1991-2001. Estimation method: GLS with cross section weights and White robust standard errors.Panel estimation by LLS and sectors and with 34 industry Fixed Effects.Level of significance: ***= 1%; **= 5%; *= 10%.Note: we have excluded local industry with zero employee in 1991 or 2001
Fichier image/jpeg, 84k
Titre Table 3. Econometric results for sectors
Légende Dependent variable: employment growth in the local industy. Annual average 1991-2001 - Cross-section estimation by LLS - Level of significance: ***= 1%; **= 5%; *= 10%Estimation Method: OLS-W: Ordinary Least Squares Estimation-White robust Standard Error, GM-error: General method of moments, IV-lag: Instrumental Variables; constant is includedNote: we have excluded local industry with zero employee in 1991 or 2001
Fichier image/jpeg, 276k
Titre Variables description and sources
Fichier image/jpeg, 90k
Haut de page

Pour citer cet article

Référence papier

Raffaele Paci et Stefano Usai, « Agglomeration economies, spatial dependence and local industry growth », Revue d'économie industrielle, 123 | 2008, 87-109.

Référence électronique

Raffaele Paci et Stefano Usai, « Agglomeration economies, spatial dependence and local industry growth », Revue d'économie industrielle [En ligne], 123 | 3e trimestre 2008, document 7, mis en ligne le 15 septembre 2010, consulté le 27 mars 2017. URL : ; DOI : 10.4000/rei.3917

Haut de page


Raffaele Paci

University of Cagliari and Crenos

Stefano Usai

University of Cagliari and Crenos

Haut de page

Droits d'auteur

© Revue d’économie industrielle

Haut de page